Prepare for a multi-AI world

Wi-Fi passwords are usually nondescript but can also be rather revealing. At last week’s annual VMware Explore conference in Barcelona, the key to connecting was the word “multi-cloud”.

It also turned out to be the key to the dominant underlying thinking behind most of the new strategies and products unveiled at the event by VMware, the world’s leading provider of software systems for managing cloud computing. The company is currently in the throes of being acquired by chip-design giant Broadcom for $69-billion, underlining its value in the market.

Much of this value was built on foreseeing the need for virtualisation of hardware and virtual machines – VMs, hence the company name – which became the heart of data centres that make cloud computing possible. It’s an industry dominated by cloud giants like Amazon Web Services, Microsoft, Google and Oracle, each of which wants to be the cloud of choice for enterprises large and small.

However, reality looks little different, says Kit Colbert, chief technology officer of VMware, who marked his 20th year with the company this year. Few businesses, he says, will use a single cloud provider.

_{Kit Colbert, VMware chief technology officer. Pic by Arthur Goldstuck}

“The future is going to be multiplayer, and you’re going to have multiple clouds,” he told us last week. “You’re going to have stuff on-premise and stuff at the edge. And so, you’ve got to think about how you architect systems to support that environment.

“We’re building this multi-cloud platform across the whole offering. It is like what we’re doing with artificial intelligence (AI). We’re trying to build AI in that same sort of cross-cloud way. So we can deliver these AI capabilities to wherever the workload is and for every customer.”

The one issue that no one is debating is whether AI will play a major role in the multi-cloud future. The highlight of the conference was VMware’s announcement of Private AI, an enterprise-ready AI framework it has developed with chipmaker Intel, led by former VMware CEO Pat Gelsinger.

“It’s clear that AI has business value, but we’re just at the starting point. Over the next few years, we’re going to see tremendous amounts of adoption across all areas. The thing with generative AI compared to traditional AI is that it is much more democratised. You don’t need to be an expert. Most of the time, you’re taking a general purpose model off the shelf.

“Where we’re going today is that you essentially have two worlds: it’s either doing little work with very little flexibility, or a lot of work giving you a lot of flexibility, but there’s no real middle ground. What we’re trying to do with Private AI is to move into that middle ground.

“The bigger challenge is, how do you put together all these different AI models in such a way that they work well together? You have your foundation model; you have a set of models on the front end for pre-processing the prompt; we can do things like retrieval, and augmented generation. And then you’re going to have AI models on the back end to do your classification, looking for hate speech and the sorts of terrible things you don’t want in the market.

“What we see is everyone who’s building their own thing is doing that same sort of stuff. So there’s a huge opportunity in the space to define what that framework looks like, to have it standardised.

“One of the big debates right now we’re seeing in the industry is whether you want to have one giant model, or a set of smaller models maybe targeted to an HR use case a customer success case, or general purpose. You’re going to see a lot of these smaller models and a ton of innovation coming out it.”

* Arthur Goldstuck is founder of World Wide Worx and editor-in-chief of Gadget.co.za. Follow him on Twitter and Instagram on @art2gee